Correlation Filter for Object Tracking Method Based on Multi-Template Update

Aimed at the current tracking algorithm such as object occlusion, severe deformation, motion blur and background confusion, a tracking method based on multiple template updates is proposed to improve the robustness of the algorithm. First, a response graph quality evaluation index is proposed to evaluate the reliability of the tracking result of the current frame. When the tracking result is unreliable, the model update is stopped immediately, and the tracker can find the object again when the object reappears. However, the indicator will always remain within a reliable range when the object is continuously blocked. At this time, if you stop the update tracking of the model, it will drift due to lack of information. In order to solve the above problems, the algorithm in this chapter adopts a multi-template tracking strategy—adding several additional filters to track the object. The proposed algorithm is compared with several recent state-of-the-art tracking algorithms on OTB100 benchmark datasets (online object tracking benchmark). Especially, the pro-posed algorithm greatly improves its basic algorithm in AUC and Precision on some complex environments of partial occlusion, severe deformation, motion blur, background clutter and illumination variation, which has a better tracking performance.